The 2017–March 2020 Pre-Pandemic Data Files for this study are sourced from the National Health and Nutrition Examination Survey (NHANES), which is an ongoing series of studies designed to assess the health and nutritional status of the civilian noninstitutionalized U.S. population. The NHANES data collection involves interviews and physical examinations, covering a wide range of health and nutrition-related topics. The data is critical for the assessment of public health and nutrition and aids in the development of health policies and research. Special weighting measures have been applied to the combined dataset to ensure that it reflects the population and health trends accurately for nationally representative estimates.
The survey utilizes a multistage probabilistic sampling design. In the first stage, PSUs are selected based on stratification by various health and demographic characteristics. Each PSU is typically a county or a group of contiguous counties.
States are categorized into health groups according to health index values. PSUs are further divided into major strata within each health group, which are determined by the urban-rural population distribution and other locality characteristics.
The 2017-2018 and 2019-March 2020 cycles were based on different 4-year sample designs, with PSUs selected annually from their respective strata.
The 2017-March 2020 pre-pandemic data file includes PSUs from the 2017-2018 cycle and 2019-March 2020 data collection. The combined dataset was calibrated to ensure national representativeness according to the 2015-2018 sample design, accounting for changes in population size and other characteristics that determine major stratum membership.
We first import three files by creating the data import function
(import_df) that reads the data from an .XPT file format. The four
datasets that were merged were: Demographic Variables and Sample Weights
(P_DEMO), Body Measures (P_BMX), Physical
Activity (P_PAQ) andDiet Behavior & Nutrition
(P_DBQ). Using a full join on the ‘SEQN’ column to merge
the datasets. Then, we select the specific column that are meaningful in
the process of analysis: SEQN, RIAGENDR,
RIDAGEYR, DMDMARTZ, INDFMPIR,
RIDRETH3, DMDEDUC2, PAD680,
BMXBMI, DBD900, and DBD910.
Next, we did the data filtering by filter out specific values from
various column (e.g., 77, 99, .,
7777, 9999). This step removed the missing,
unknown, and null value from the dataset and only keep those meaningful
values in the dataset.
Furthermore, we rename the columns to more descriptive names like ‘gender’, ‘age’, ‘marital_status’, etc. After this step, we use the mutate function to transform and categorize several columns:
gender: Categorizing and converting to a factor.marital_status, race,
education: Categorizing different demographic details.obese: Creating a new variable based on BMI,
categorizing individuals as ‘normal’ or ‘obese’.The variables in the Final dataset are listed below. -
SEQN : Respondent sequence number - gender:
Female or male - age: Age in years at screening -
marital_status: never married, married,
widowed/divorced/separated - income_to_poverty: ratio of
family income to poverty - race: Mexican American, White,
Black, Other Hispanic - bmi: Body Mass Index (kg/m**2) -
freq_fast_food: number of meals from fast food or pizza
place - freq_frozen: Numbers of frozen meals/pizza in past
30 days - sedentary_activity: Minutes sedentary activity -
obese: normal, obese
The process effectively cleans and transforms the data, preparing it for analysis related to obesity. The use of case_match and case_when for categorizing data is particularly efficient, as it allows for more readable and interpretable data. This structured approach ensures the dataset is clean, well-organized, and ready for further statistical analysis or visualization.